A method and system for matching ground elements between multiple batches of relative maps

Abstract

The present invention relates to a method and system for matching ground elements between multiple batches of relative maps. The matching method includes: selecting map frames of relative maps and their key points, dividing each map frame centered on the key points, and outputting the frame ranges of the map frames; obtaining the reliable ground feature types and corresponding quantities included in the ground object within the frame range, and determining the key frames in the map frame according to the included ground feature types and corresponding quantities; comparing the ground objects to match the neighboring key frames, and outputting the association relationships of the neighboring key frames; after adjusting the global association result, outputting the ground object association result; performing matching based on semantic elements, being more tolerant to data errors, more in line with the actual production environment, easy to debug, convenient to control, having good interpretability, high robustness, directly expressing data features, and suitable for use in the real environment of large-scale data.

Description

Technical Field

[0001] The present invention relates to the field of high-precision map production, and particularly to a method and system for matching ground elements between multi-batch relative maps. Background Art

[0002] In the field of high-precision map production, the fusion of multi-batch relative maps is often involved. A relative map refers to a local map constructed by a camera and an IMU (Inertial Measurement Unit) when a single vehicle moves, and is registered to an absolute position through a low-precision GPS reference. The absolute accuracy of ground features is about 1 - 10 m. A single-batch relative map refers to the result of a single vehicle running once to form a map. A multi-batch relative map refers to the result of multiple vehicles running multiple times to form a map, which is actually multiple single-batch relative maps.

[0003] The matching of ground features between multi-batch relative maps, that is, estimating which objects in different batch relative maps belong to the same actual physical object. For example, a vehicle passes by a certain place 4 times and sees traffic lights 3 times. Whether these traffic lights seen at different times correspond to the same traffic light in the real physical world. This matching method will determine which ground feature objects in the relative maps seen multiple times belong to the same real object in the physical world and give a relatively reliable association result.

[0004] Taking the low-precision GPS as a reference benchmark, the absolute accuracy of relative maps is generally within 10 meters. In the process of fusing multi-batch relative maps, the matching process of ground elements is the most critical. After the ground elements of different batch maps can be correctly matched and associated, the subsequent optimization can be continued. However, the matching method of ground elements is often difficult to obtain reliable and stable results due to the data collection environment, data quality, and accuracy changes.

[0005] Conventional matching methods for ground elements in different batches include image semantic descriptor matching in SLAM. However, in actual situations, the descriptor extraction methods of these methods are mostly obtained by experience, and the results do not always conform to the actual environment. In the actual production process, a fast, reliable, and controllable method is needed. Summary of the Invention

[0006] The present invention addresses the technical problems existing in the prior art and provides a method and system for matching ground elements between multiple batches of relative maps. The matching is based on semantic elements, with higher tolerance for data errors, more in line with the actual production environment, easier to debug, convenient to control, good interpretability, high robustness, and direct expression of data features, suitable for use in the real environment of large-scale data. The matching result is completely determined by the input data. The larger the data volume and the more cross-validation, the closer the matching result is to the actual situation. This solution makes full use of information, can mine the true matching information from the data, effectively reduces the intervention of complex rules, greatly simplifies the complexity of the system, and is conducive to online operation in the real environment.

[0007] According to the first aspect of the present invention, there is provided a method for matching ground elements between multiple batches of relative maps, including: Step 1, selecting the map frames and their key points of the relative maps, dividing each of the map frames centered on the key points, and outputting the frame ranges of the map frames;

[0008] Step 2, obtaining the reliable ground object types and their corresponding quantities included in the frame ranges, and determining the key frames in the map frames according to the included ground object types and their corresponding quantities;

[0009] Step 3, comparing the ground objects to match the neighboring key frames, outputting the association relationships of the neighboring key frames; after adjusting the global association result, outputting the ground object association result.

[0010] Based on the above technical solution, the present invention can also be improved as follows.

[0011] Optionally, the process of selecting the map frames and their key points in Step 1 includes:

[0012] Along the trajectory direction of the relative map, selecting trajectory key points at a set distance interval, and establishing the frame ranges of regions of a set size centered on the key points;

[0013] The region of the set size is a square, and the range of the side length of the square is 40 ± 5 meters.

[0014] Optionally, in Step 2, it is determined whether the position of the ground object type is reliable according to the frequency of occurrence of the ground object type in each map frame, and the lower the frequency, the more reliable the position.

[0015] Optionally, in Step 2, the reliable ground object types and whether the type exists in the frame are represented in the form of bag-of-words encoding, including:

[0016] Select the feature objects within the frame range, and construct a feature vector including the types of feature objects included in the frame range and whether each type exists in the frame. The feature vector is one-hot encoded, that is, a column vector of 1*N, where N represents the number of types of feature objects. Each element in the vector represents whether the feature objects of each type arranged in sequence exist in the frame in turn.

[0017] Optionally, the process of determining the key frames in the map frame in step 2 includes:

[0018] Take the map frame containing the specified key type of feature objects as the key frame, and obtain the position information of the feature objects of each key frame.

[0019] Optionally, the process of outputting the association relationship of neighboring key frames in step 3 includes:

[0020] Step 301, traverse the key frames, and obtain all neighboring key frames Q whose distance from the current key frame P is less than the specified distance;

[0021] Step 302, separate the feature objects in the current key frame P and the neighboring key frame Q by type;

[0022] Step 303, calculate the minimum matching distance of key frames of the same type of feature objects between the current key frame P and the neighboring key frame Q, and use the minimum matching distance of key frames to represent the conditional probability of the matching relationship between the current key frame P and the neighboring key frame Q;

[0023] Step 304, filter out invalid neighboring key frame association relationships according to the real environment.

[0024] Optionally, the calculation method of the minimum matching distance of key frames in step 303 is:

[0025] Step 30301, calculate the minimum distance between the current key frame P and the neighboring key frame Q for the same type of feature objects as the distance between feature objects;

[0026] Step 30302, calculate the average value of the distances between feature objects within the class as the distance between feature objects within the class;

[0027] Step 30303, calculate the key frame distance as the weighted sum of the distances between feature objects within each class, and the weight value is determined by the occurrence frequency of each type of feature object;

[0028] Step 30304, translate the feature object in the current key frame P to a certain feature object of the same type in the neighboring key frame Q to obtain the key frame P', and calculate the key frame distance between the key frame P' and the neighboring key frame Q as the key frame matching distance of the feature object;

[0029] Step 30305: Calculate the key-frame matching distance for the class as the minimum value of the key-frame matching distances of each feature object of any class in the current key frame P and the neighboring key frame Q.

[0030] Step 30306: Take the minimum value of the key-frame matching distances of all the classes in the current key frame P and the neighboring key frame Q as the minimum key-frame matching distance.

[0031] Optionally, in step 304, filtering out invalid neighboring key-frame association relationships according to the real environment includes:

[0032] Define the association relationships with key-frame matching distances exceeding the set minimum threshold as invalid relationships.

[0033] When there is one key frame corresponding to multiple key frames in the association relationship, define the association relationship corresponding to the minimum key-frame matching distance among these key frames as the valid association relationship, and the rest as invalid association relationships.

[0034] According to the second aspect of the present invention, there is provided a matching system for ground elements between multiple batches of relative maps. The matching system includes: a frame range output module, a key frame acquisition module, and an association result output module;

[0035] The frame range output module is configured to select map frames of the relative map and their key points, divide each of the map frames centered on the key points, and output the frame ranges of the map frames.

[0036] The key frame acquisition module is configured to acquire the reliable feature types and corresponding quantities of the feature objects included in the frame range, and determine the key frames in the map frames according to the included feature types and corresponding quantities.

[0037] The association result output module is configured to match the feature objects to the neighboring key frames and output the association relationships of the neighboring key frames; after adjusting the global association result, output the feature association result.

[0038] A method and system for matching ground elements between multiple batches of relative maps provided by the present invention use semantic information as the data source instead of the original image information. Semantic information is often expressed in vector shapes such as points and lines. While greatly reducing the stored information, the relative relationships between ground objects are still retained, and the utilization rate of data information is relatively high. In actual implementation, through key frame extraction, the data volume is further reduced. By comparing two key frames pairwise, the best matching effect between different ground elements is obtained, and then all the matching results are summarized for cross-validation. The more cross-validations are performed, the higher the matching confidence. The matching result is completely determined by the input data. The larger the data volume and the more cross-validations are performed, the closer the matching result is to the actual situation. This solution makes full use of information, can mine the true matching information from the data, effectively reduces the intervention of complex rules, greatly simplifies the complexity of the system, has a higher tolerance for data errors, and is conducive to running online in a real environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 is a flowchart of a method for matching ground elements between multiple batches of relative maps provided by the present invention;

[0040] Figure 2 is a schematic diagram of the key frames extracted in an embodiment of the present invention;

[0041] Figure 3 is a schematic diagram of the global association effect in an embodiment of the present invention;

[0042] Figure 4 is a structural block diagram of a system for matching ground elements between multiple batches of relative maps provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0043] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0044] Figure 1 is a flowchart of an embodiment of a method for matching ground elements between multiple batches of relative maps provided by the present invention, Figure 1 where the left rectangular boxes are the respective processes, and the right cylindrical boxes are the input and output data of the corresponding processes. As can be seen from Figure 1 the matching method includes:

[0045] Step 1, select the map frames and their key points of the relative maps, and divide each map frame centered on the key points to output the frame range of the map frame.

[0046] The purpose of establishing the frame range of the map frame is to convert the relative maps with different ranges into map frames of a fixed size. These map frames form the basic unit for comparing ground object pairs. Otherwise, the comparison process of relative maps with different ranges is difficult.

[0047] Step 2: Obtain the reliable feature types and corresponding quantities of the feature objects included within the frame range, and determine the key frames in the map frame according to the included feature types and corresponding quantities.

[0048] Step 3: Compare the feature objects to match the neighboring key frames, and output the association relationships of the neighboring key frames; after adjusting the global association results, output the feature association results.

[0049] A method for matching ground features between multiple batches of relative maps provided by the present invention performs matching based on semantic features, has higher tolerance for data errors, is more in line with the actual production environment, is easy to debug, convenient to control, has good interpretability, and high robustness, makes a direct expression of data features, and is suitable for use in the real environment of large-scale data; the matching result is completely determined by the input data. The larger the data volume and the more cross-validations, the closer the matching result is to the actual situation. This solution makes full use of information, can mine the true matching information from the data, effectively reduces the intervention of complex rules, greatly simplifies the complexity of the system, and is conducive to online operation in the real environment.

[0050] Embodiment 1

[0051] Embodiment 1 provided by the present invention is an embodiment of a method for matching ground features between multiple batches of relative maps provided by the present invention. Combining Figure 1 it can be seen that the embodiments of this matching method include:

[0052] Step 1: Select the map frames and their key points of the relative maps, divide each map frame centered on the key points, and output the frame range of the map frame.

[0053] In a possible embodiment, the process of selecting the map frames and their key points in Step 1 includes:

[0054] Along the trajectory direction of the relative map, select the trajectory key points at a set distance interval, and establish a frame range of a set size area centered on the key points.

[0055] The set size area is a square, and the range of the side length of the square is 40 ± 5 meters.

[0056] In specific implementation, the set distance interval can be 20 meters, and the square of the set size area can be 40 × 40 meters. The 40-meter frame range is based on the fact that the number of feature objects within this range is neither too many nor too few, which can not only ensure a certain number of feature objects within the frame to be able to interact with each other, but also prevent too many feature objects from causing excessive data volume and complex calculations. In the subsequent comparison of two-by-two map frames, the time complexity of the algorithm is O(n 2) In the case where the complexity cannot be reduced, controlling the number of n can effectively control the running time of the algorithm.

[0057] The ranges of each map frame should overlap to a certain extent to avoid missing the association of edge feature objects.

[0058] Step 2: Obtain the reliable feature types and corresponding quantities of the feature objects contained within the frame range, and determine the key frames in the map frame based on the contained feature types and corresponding quantities.

[0059] As Figure 2 shown is a schematic diagram of the key frames extracted provided by the embodiment of the present invention. Figure 2 In it, the picture is the result of the segmentation of the key frames of a single batch, and there will be key frames only if there are available features. The lines running through from top to bottom represent the trajectories; the light-colored dots represent traffic lights; the squares filled with diagonal lines represent zebra crossings; the horizontal lines represent stop lines; the vertical squares represent direction arrows, the dark-colored dots represent traffic signs, and the dotted-line boxes are the key frame ranges.

[0060] In a possible embodiment, in step 2, the reliability of the position of the feature type is determined according to the frequency of occurrence of the feature type in each map frame. The lower the frequency, the more reliable the position.

[0061] During the selection process of available feature types, those with a relatively low frequency of occurrence can be used as feature objects for reference positioning, such as traffic signs and zebra crossings. If such feature objects exist in every map frame and have a relatively high frequency of occurrence, such as lane lines and poles, they are not used as available feature objects.

[0062] In a possible embodiment, in step 2, the reliable feature types and corresponding quantities are represented in the form of bag-of-words encoding, including:

[0063] Select the feature objects within the frame range, and construct a feature vector of the feature types contained in this frame range and whether this type exists. The feature vector is based on one-hot encoding, that is, a 1*N column vector, where N represents the number of feature types, and each element is in turn whether the feature objects of each type arranged in order exist. 1 represents existence, and 0 represents non-existence.

[0064] The bag-of-words encoding only describes the presence or absence of a certain type, without making more detailed expressions. This is because when generating a relative map, due to environmental occlusion, the actual physical objects are not always visible. Through simple bag-of-words expressions, key frames can be quickly pre-screened. Borrowing the definition of bag-of-words in natural language processing, the encoding method here can be called bag-of-words encoding based on ground object semantics. In a specific implementation, an example of bag-of-words encoding can be: selecting traffic lights, traffic signs, stop lines, direction arrows, and zebra crossings as available ground object types. If a certain frame has 2 direction arrows, 1 stop line, and 1 zebra crossing, then the feature vector of this frame is [0, 0, 1, 1, 1].

[0065] Ground object expression methods: Ground objects have expression methods such as bounding box rectangles, contour polygons, and center points. Due to the relatively large observation errors of relative maps, in order to improve fault tolerance, the ground object expression method of the present invention is the center point.

[0066] In one possible embodiment, the process of determining key frames in map frames in step 2 includes:

[0067] Regarding map frames containing ground object objects of set key types as key frames, and obtaining the position information of ground object objects that can be used in each key frame.

[0068] In a specific implementation, according to the bag-of-words encoding, map frames containing ground object objects of key types are obtained as key frames. Extracting key frames can effectively reduce the data volume. For frames without available ground object objects, the information content is small and does not need to participate in the calculation.

[0069] Step 3, compare ground object objects to match neighboring key frames, and output the association relationship of neighboring key frames; after adjusting the global association result, output the ground object association result.

[0070] In one possible embodiment, the process of outputting the association relationship of neighboring key frames in step 3 includes:

[0071] Step 301, traverse the key frames to obtain all neighboring key frames Q whose distance from the current key frame P is less than a set distance (for example, 10 meters).

[0072] Step 302, separate the ground object objects in the current key frame P and the neighboring key frames Q by type.

[0073] Step 303, calculate the minimum matching distance of key frames of the same type of ground object objects between the current key frame P and the neighboring key frames Q, and use the minimum matching distance of key frames to represent the conditional probability of the matching relationship between the current key frame P and the neighboring key frames Q.

[0074] In one possible embodiment, the calculation method of the minimum matching distance of key frames in step 303 is:

[0075] Step 30301: Calculate the ground object distance as the shortest distance between the same type of ground object in the current key frame P and the neighboring key frame Q.

[0076] Step 30302: Calculate the in-class ground object distance as the average value of the distances of all ground objects within the class.

[0077] Step 30303: Calculate the key frame distance as the weighted sum of the in-class ground object distances, where the weights are determined by the occurrence frequencies of various types of ground objects.

[0078] Step 30304: Translate the ground object in the current key frame P to a certain ground object of the same type in the neighboring key frame Q to obtain the key frame P'. Calculate the key frame distance between the key frame P' and the neighboring key frame Q as the key frame matching distance of this ground object.

[0079] Step 30305: Calculate the key frame matching distance of the class as the minimum value of the key frame matching distances of each ground object of any type in the current key frame P and the neighboring key frame Q.

[0080] Step 30306: Take the minimum value of the key frame matching distances of all classes in the current key frame P and the neighboring key frame Q as the minimum key frame matching distance. The matching relationship corresponding to this minimum key frame matching distance is the preliminary result of pairwise key frame matching.

[0081] The purpose of finding the minimum key frame matching distance is: Given the current key frame P and the neighboring key frame Q, find the maximum value of the conditional probabilities of various matching relationships between the current key frame P and the neighboring key frame Q. This maximum probability value is equivalent to the minimum value of the key frame matching distance.

[0082] Step 304: Filter out invalid neighboring key frame association relationships according to the real environment.

[0083] In a possible embodiment, filtering out invalid neighboring key frame association relationships in Step 304 includes:

[0084] Locate the association relationships with key frame matching distances exceeding the set minimum threshold as invalid relationships.

[0085] Since the association relationships are taken according to the shortest distance, there may be a situation where one object corresponds to multiple objects. Duplicate corresponding situations should be filtered out. The specific method is: When there is an association relationship where one key frame corresponds to multiple key frames, determine the association relationship corresponding to the minimum key frame matching distance among these key frames as the valid association relationship, and the rest as invalid association relationships.

[0086] In a possible embodiment, the method for adjusting the global association result in Step 3 is to filter the association relationships of key frames according to global constraints. The filtering conditions include:

[0087] 1) Global physical distance filtering: According to the characteristics of the input data, the global associated feature objects should be within a set distance (e.g., 10 meters).

[0088] 2) The same feature object should not point to multiple feature objects in the same batch; when the same feature object points to multiple feature objects in the same batch, sort them according to the global physical distance and retain the associated relationship corresponding to the minimum distance.

[0089] 3) Merge the forward and reverse associated relationships between feature objects and remove duplicate corresponding phenomena.

[0090] 4) Graph search verification: Using feature objects as vertices and associated relationships as edges to form a graph network, find the indirect associated relationships of feature objects. If there are no conflicts in the indirect associated relationships, or even form a closed-loop associated relationship, it means the highest confidence. If there are conflicting indirect associated relationships, take the relationship with the minimum average distance as the valid associated relationship and discard other associated relationships.

[0091] As Figure 3 shown in the schematic diagram of the effect of global association provided by the embodiment of the present invention, Figure 3 in which the arrowed connection lines running from bottom to top through the picture are trajectories; the 4 rectangular frames in the upper part are zebra crossings seen from this place; the 3 horizontal lines in the middle are stop lines (the upper 2 coincide); the many vertical rectangles in the lower part are direction arrows. The black arrow in the middle marks the associated direction, and the double-headed arrow indicates a two-way association. The association between the stop line and the direction arrow in the figure has formed a closed loop.

[0092] Embodiment 2

[0093] Embodiment 2 provided by the present invention is an embodiment of a matching system for ground elements between multiple batches of relative maps provided by the present invention. Figure 4 It is a structural diagram of a matching system for ground elements between multiple batches of relative maps provided by an embodiment of the present invention. As shown in Figure 4 it can be seen that the embodiment of this matching system includes: a frame range output module, a key frame acquisition module, and an associated result output module.

[0094] The frame range output module is used to select the map frames and their key points of the relative maps, divide each map frame centered on the key points, and output the frame ranges of the map frames.

[0095] The key frame acquisition module is used to obtain the reliable ground feature types and the corresponding quantities in the feature objects included in the frame range, and determine the key frames in the map frames according to the included ground feature types and the corresponding quantities.

[0096] The associated result output module is used to compare the feature objects to match the neighboring key frames and output the association relationships of the neighboring key frames; after adjusting the global association results, it outputs the feature association results.

[0097] It can be understood that a matching system for ground features between multiple batches of relative maps provided by the present invention corresponds to the matching methods for ground features between multiple batches of relative maps provided in the foregoing embodiments. The relevant technical features of the matching system for ground features between multiple batches of relative maps can refer to the relevant technical features of the matching methods for ground features between multiple batches of relative maps, which will not be elaborated here.

[0098] A matching method and system for ground features between multiple batches of relative maps provided by an embodiment of the present invention solve the problem that it is difficult to match ground features between multiple batches of relative maps. The data source is semantic information, rather than the original image information. Semantic information is often expressed in vector shapes such as points and lines. While greatly reducing the stored information, it still retains the relative relationships between features, and the utilization rate of data information is relatively high; in actual implementation, the amount of data is further reduced through key frame extraction. By comparing two key frames pairwise, the best matching effect between different ground features is obtained, and then all the matching results are summarized for cross-validation. The more cross-validations, the higher the matching confidence; the matching result is completely determined by the input data. The larger the amount of data and the more cross-validations, the closer the matching result is to the actual situation. This solution makes full use of information, can effectively reduce the intervention of complex rules, greatly simplifies the complexity of the system, has a higher tolerance for data errors, and is conducive to being put into operation in a real environment.

[0099] It should be noted that in the above embodiments, the descriptions of each embodiment have their own emphases. For the parts not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0100] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can be implemented in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can be implemented in the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0101] The present invention will be described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in one block or multiple blocks.

[0102] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implement the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in one block or multiple blocks.

[0103] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in one block or multiple blocks.

[0104] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications once they know the basic inventive concept. Therefore, the appended claims are intended to be construed as including the preferred embodiments and all changes and modifications falling within the scope of the present invention.

[0105] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. A method for matching ground elements between multiple batches of relative maps, characterized in that, The method includes: Step 1: Select a map frame and its key points relative to the map, divide each of the map frames centered on the key points, and output the frame range of the map frames. Step 2: Obtain the reliable ground feature types and corresponding quantities included in the ground feature objects within the frame range, and determine the key frames in the map frames according to the included ground feature types and corresponding quantities. Step 3: Compare the ground feature objects to match the neighboring key frames, output the association relationship of the neighboring key frames; after adjusting the global association result, output the ground feature association result. The process of outputting the association relationship of the neighboring key frames in Step 3 includes: Step 301: Traverse the key frames to obtain all neighboring key frames Q whose distance from the current key frame P is less than the set distance. Step 302: Separate the ground feature objects within the current key frame P and the neighboring key frame Q by type. Step 303: Calculate the minimum matching distance of the key frames of the same type of ground feature objects between the current key frame P and the neighboring key frame Q, and use the minimum matching distance of the key frames to represent the conditional probability of the matching relationship between the current key frame P and the neighboring key frame Q. Step 304: Filter out invalid association relationships of the neighboring key frames according to the real environment. The calculation method of the minimum matching distance of the key frames in Step 303 is: Step 30301: Calculate the ground feature distance as the shortest distance between the same type of ground feature objects in the current key frame P and the neighboring key frame Q. Step 30302: Calculate the within-class ground feature object distance as the average value of all the ground feature object distances within the class. Step 30303: Calculate the key frame distance as the weighted sum of the within-class ground feature object distances, and the weights are determined by the occurrence frequencies of various types of ground feature objects. Step 30304: Translate the ground feature object in the current key frame P to a certain ground feature object of the same type in the neighboring key frame Q to obtain the key frame P’, and calculate the key frame distance between the key frame P’ and the neighboring key frame Q as the key frame matching distance of this ground feature object. Step 30305: Calculate the key frame matching distance of the class as the minimum value of the key frame matching distances of each ground feature object of any type between the current key frame P and the neighboring key frame Q. Step 30306: Take the minimum value of the key frame matching distances of all the classes between the current key frame P and the neighboring key frame Q as the minimum matching distance of the key frames.

2. The matching method according to claim 1, wherein The process of selecting a map frame and its key points relative to the map in Step 1 includes: Along the trajectory direction of the relative map, select trajectory key points at set distance intervals, and centered on the key points, establish the frame range of a region of set size. The region of set size is a square, and the range of the side length of the square is 40 ± 5 meters.

3. The matching method according to claim 1, wherein In Step 2, determine whether the position of the ground feature type is reliable according to the frequency of occurrence of the ground feature type in each map frame, and the lower the frequency, the more reliable the position.

4. The matching method according to claim 1, characterized in that, In Step 2, represent the reliable ground feature types and whether the ground feature of this type exists in this frame in the form of bag-of-words encoding, including: Select the feature objects within the frame range, and construct a feature vector including the types of feature objects included in the frame range and whether each type exists in the frame. The feature vector is one-hot encoded, that is, a 1 * N column vector, where N represents the number of types of feature objects. Each element in the feature vector represents whether the feature objects of each type arranged in sequence exist in turn.

5. The matching method according to claim 1, characterized in that, The process of determining the key frames in the map frame in step 2 includes: Regarding the map frame containing the specified key type of feature objects as key frames, and obtaining the position information of the feature objects of each key frame.

6. The matching method according to claim 1, characterized in that The filtering of invalid neighboring key frame association relationships according to the real environment in step 304 includes: Defining the association relationships with the key frame matching distance exceeding the set minimum threshold as invalid relationships; When there is an association relationship where one key frame corresponds to multiple key frames, defining the association relationship corresponding to the minimum key frame matching distance among these key frames as the valid association relationship, and the rest as invalid association relationships.

7. The matching method according to claim 1, characterized in that The method for adjusting the global association result in step 3 is to filter the association relationships of the key frames according to global constraints. The filtering conditions include: The global associated feature object distances should be within the set distance; The same feature object should not point to multiple feature objects of the same batch; when the same feature object points to multiple feature objects of the same batch, sort them according to the global physical distance, and retain the association relationship corresponding to the minimum distance; Merge the forward and reverse association relationships between feature objects, and remove the duplicate corresponding phenomena Taking the feature objects as vertices and the association relationships as edges to form a graph network, searching for the indirect association relationships of the feature objects. If there are conflicting indirect association relationships, taking the relationship with the minimum average distance as the valid association relationship and abandoning other association relationships.

8. A matching system for ground elements between multiple batches of relative maps, characterized in that, The matching system includes: a frame range output module, a key frame acquisition module, and an association result output module; The frame range output module is used to select the map frames and their key points relative to the map, divide each map frame centered on the key points, and output the frame ranges of the map frames; The key frame acquisition module is used to obtain the reliable types and corresponding quantities of the feature objects included in the frame range, and determine the key frames in the map frame according to the included types and corresponding quantities of the feature objects; The association result output module is used to compare the feature objects to match the neighboring key frames, and output the association relationships of the neighboring key frames; after adjusting the global association result, output the feature object association result; The process of the association result output module outputting the association relationships of the neighboring key frames includes: Step 301, traverse the key frames to obtain all neighboring key frames Q whose distance from the current key frame P is less than the set distance; Step 302, separate the feature objects in the current key frame P and the neighboring key frame Q by type; Step 303, calculate the minimum key frame matching distance of the same type of feature objects between the current key frame P and the neighboring key frame Q, and use the minimum key frame matching distance to represent the conditional probability of the matching relationship between the current key frame P and the neighboring key frame Q; Step 304, filter the invalid neighboring key frame association relationships according to the real environment; The calculation method of the minimum matching distance of the key frame in step 303 is as follows: Step 30301: Calculate the ground object distance as the nearest distance between the same type of ground object in the current key frame P and the neighboring key frame Q; Step 30302: Calculate the in-class ground object distance as the average value of all the ground object distances within the class; Step 30303: Calculate the key frame distance as the weighted sum of the in-class ground object distances, where the weights are determined by the occurrence frequencies of various types of ground objects; Step 30304: Translate the ground object in the current key frame P to a certain ground object of the same type in the neighboring key frame Q to obtain the key frame P’, and calculate the key frame distance between the key frame P’ and the neighboring key frame Q as the key frame matching distance of this ground object; Step 30305: Calculate the key frame matching distance of the class as the minimum value of the key frame matching distances of each ground object of any type in the current key frame P and the neighboring key frame Q; Step 30306: Take the minimum value of the key frame matching distances of all the classes in the current key frame P and the neighboring key frame Q as the minimum matching distance of the key frame.

Images